Intel strikes back

[hg2]

Do not click if u are apple die-hard fans....
https://www.pcworld.com/article/3606592/intel-benchmarks-say-apples-m1-isnt-faster.html

 

[TrueBeliever_jh]

Is this the cherry picked benchmark using their top of the line i7 against apple’s 1st gen entry m1 chip?

So who’s th target audience of these benchmarks? MacBook buyers?

 

[ragnarok95]

Jelly posting from intel since Apple cut them off on their CPU and their modem. Who to blame? Intel. Apple is going the right way of making their own SoC. All i can say is they will get better.

Btw, i'm not a apple fan boy and i will never buy a mac running mac OS.

 

[NoMotherTeachSiboh]

intel claims benchmarks not important...yet uses benchmarks

 

[hg2]

Is this the cherry picked benchmark using their top of the line i7 against apple’s 1st gen entry m1 chip?

So who’s th target audience of these benchmarks? MacBook buyers?

In transcoding, apps compatibility and gaming domain, no way M1 can beat intel+Nvidia combo... common sense i7+3090 win hands down...

On the other hand, in terms of fan noise and heat generated M1 wins hands down.....

Battery hours tie... very convincing....

 

[TrueBeliever_jh]

In transcoding, apps compatibility and gaming domain, no way M1 can beat intel+Nvidia combo... common sense i7+3090 win hands down...

On the other hand, in terms of fan noise and heat generated M1 wins hands down.....

Battery hours tie... very convincing....

once again... shouldnt they be aiming at their contemporary~?

let me help you out...

psst psst *AMD*


their target sure is real low aiming at some ARM CPU~

 

[enenyi]

once again... shouldnt they be aiming at their contemporary~?

let me help you out...

psst psst *AMD*


their target sure is real low aiming at some ARM CPU~

Intel targeting AMD is short-term and short-sighted.
Intel targeting Apple (and ARM in general) is looking further ahead.
Coz if ARM becomes more popular, both Intel and AMD will lose...together.

 

[kenzi87]

Intel change boss mah, confirm got direction change

 

[hg2]

once again... shouldnt they be aiming at their contemporary~?

let me help you out...

psst psst *AMD*


their target sure is real low aiming at some ARM CPU~

In Laptop arena, intel have an army of engineers optimizing and building chipsets drivers... AMD although gaining ground in terms of raw power, many areas they are still lacking. e.g, battery power management, I/O... worst AMD is at the mercy of TSMC... any large scale volume increase is impossible, as it's already running at capacity..... whereas intel can any time double the productions and lower the price by 50%..... If i run intel, by simply Quadruple productions and cut price by 75%... AMD gg

 

[wwenze]

This does being up one interesting issue...

Most of us PC users have never noticed this since x86 processors are largely similarly with each other

However on the mobile CPU side, you can have CPU A being 2x as fast as CPU B for one task, but CPU B is 50% faster than CPU A for another task.

So what do you look for when looking at benchmarks and reviews? Well apparently people just buy iPhone at the end of the day.

 

[wwenze]

In Laptop arena, intel have an army of engineers optimizing and building chipsets drivers... AMD although gaining ground in terms of raw power, many areas they are still lacking. e.g, battery power management, I/O... worst AMD is at the mercy of TSMC... any large scale volume increase is impossible, as it's already running at capacity..... whereas intel can any time double the productions and lower the price by 50%..... If i run intel, by simply Quadruple productions and cut price by 75%... AMD gg

Intel? Double production and reduce price?

Hahahahahahahahah

Maybe for their 14nm++++++++++ process

Last I checked, they have given up on their 10nm gen1 for desktop usage. Better hope superfin makes it on time.

 

[wwenze]

Intel targeting AMD is short-term and short-sighted.
Intel targeting Apple (and ARM in general) is looking further ahead.
Coz if ARM becomes more popular, both Intel and AMD will lose...together.

Nvidia using ARM for their car driving systems

But Nvidia using x86 for their DGX supercomputers

One advantage that x86 still has over ARM, and this even shows up in gaming consoles as well, is that x86 has been built over the years to focus on speed to "dumb" code. And majority of code done by human is very dumb.

Dumb here just means very sequential with lots of dependencies in the lots of branching.

You will find that if you try to run the same game at zero graphics and unlimited fps, you will tend to get few hundreds to thousands FPS with x86 and far from that with ARM.

 

[enenyi]

Nvidia using ARM for their car driving systems

But Nvidia using x86 for their DGX supercomputers

One advantage that x86 still has over ARM, and this even shows up in gaming consoles as well, is that x86 has been built over the years to focus on speed to "dumb" code. And majority of code done by human is very dumb.

Dumb here just means very sequential with lots of dependencies in the lots of branching.

You will find that if you try to run the same game at zero graphics and unlimited fps, you will tend to get few hundreds to thousands FPS with x86 and far from that with ARM.

For now the dumb code is still with x86. Coz ARM has played catch-up up till now.
Moving forward there might be a possibility that dumb code will be moved over to ARM for efficiency reasons. Then most/all parallel computations can be done using GPU/FPGA. This leaves x86 in no-man's land and I can only speculate this is something that Intel is trying desperately to avoid. AMD is not even a threat if both of them will end up in the same muddy situation anyway.

 

[wwenze]

Personally I believe full parallelization can never be possible when it comes to decision making. A combination of something as a dumb overseer to assign tasks to embarassingly-parallel GPUs is likely the future. Even AI network computing doesn't help here, since we are getting bottlenecked by the signal propagation speed (which also shows itself as the memory speed wall), but x86 has had lots of features to ownself speculate the execution ownself. Although M1 has improved itself by a lot over the original ARM.

The question however is, how often do we need final results at the rate of more than 200 results per second. Once a certain amount of speed for simple code is reached, we don't really need it to be faster, while we need complex tasks to be faster but those can be parallelized. Of course, Wirth's law will manage to undo any speed advances so the arms race remains. And from personal experience, I think Wirth's law is winning. I mean look at how bad game ports can be.

 

[GBC]

Intel change boss mah, confirm got direction change

Yeah sure. The CEO's pay goes to a different pocket only..

 

[enenyi]

Personally I believe full parallelization can never be possible when it comes to decision making. A combination of something as a dumb overseer to assign tasks to embarassingly-parallel GPUs is likely the future. Even AI network computing doesn't help here, since we are getting bottlenecked by the signal propagation speed (which also shows itself as the memory speed wall), but x86 has had lots of features to ownself speculate the execution ownself. Although M1 has improved itself by a lot over the original ARM.

I don't think the dumb overseer contributes much to the parallelization except to initiate the process then consolidate the results to verify accuracy or ensure completion (think General Purpose GPU computation). Speculation is also what made x86 processors comparatively faster over the years, but introduced security flaws that can be exploited, so double-edged sword there. The penalty for undoing a speculative branch can sometimes be worse than not speculating at all, dependent on the workload in question. ARM's appeal is not in performance but in efficiency. If an x86 chip (Intel or AMD) has the same computation time on a given task as an ARM chip but requires way more power to do so, wouldn't laptop makers eventually adopt ARM chips to boost their 'battery life' marketing? Something to think about in the x86 vs ARM argument.

The question however is, how often do we need final results at the rate of more than 200 results per second. Once a certain amount of speed for simple code is reached, we don't really need it to be faster, while we need complex tasks to be faster but those can be parallelized. Of course, Wirth's law will manage to undo any speed advances so the arms race remains. And from personal experience, I think Wirth's law is winning. I mean look at how bad game ports can be.

Not sure why we don't need final results more than 200 results per second. The whole idea of parallelization is to address this requirement of higher speed, though parallelization brought more problems such as synchronization and accuracy to the table. If a given program runs for 10 minutes to achieve a task, wouldn't it be natural to want the program to run faster (i.e. under 10 minutes)? Then with faster programs, the energy consumption, however little or large, is reduced both on desktop (AC power) and laptop (battery power) environments? If taking just game development into play (game porting), game developers are likely less incentivized to rewrite code for another platform for same profits, so maybe those efficiency arguments don't apply. But for other computational purposes, wouldn't spending less money on electricity for the same compute output be ideal?